--- title: "jailbreak-evaluation vs agentdojo" type: "comparison" canonical_url: "https://www.graphcanon.com/compare/controllability-jailbreak-evaluation-vs-ethz-spylab-agentdojo" tools: ["controllability-jailbreak-evaluation", "ethz-spylab-agentdojo"] --- # jailbreak-evaluation vs agentdojo *GraphCanon updated Aug 5, 2026* ## Verdict Pick jailbreak-evaluation if jailbreak-evaluation is a Python package aimed at evaluating if AI models have been jailbroken by generating outputs that diverge from expected programming; pick agentdojo if agentDojo serves as a benchmarking environment to evaluate security attacks, like prompt injection, and defenses for Large Language Model (LLM) agents. [jailbreak-evaluation](https://arxiv.org/abs/2404.06407) reports 27 GitHub stars, 8 forks, and 0 open issues, last pushed Nov 4, 2024. [agentdojo](https://agentdojo.spylab.ai/) has 716 stars, 188 forks, and 41 open issues, last pushed Jun 2, 2026. Figures are from public GitHub metadata via [jailbreak-evaluation's repository](https://github.com/controllability/jailbreak-evaluation) and [agentdojo's repository](https://github.com/ethz-spylab/agentdojo). | | [jailbreak-evaluation](/tools/controllability-jailbreak-evaluation.md) | [agentdojo](/tools/ethz-spylab-agentdojo.md) | | --- | --- | --- | | Tagline | Python package for language model jailbreak evaluation | A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents | | Stars | 27 | 716 | | Forks | 8 | 188 | | Open issues | 0 | 41 | | Language | Python | Python | | Adopt for | jailbreak-evaluation is a Python package aimed at evaluating if AI models have been jailbroken by generating outputs that diverge from expected programming. | AgentDojo serves as a benchmarking environment to evaluate security attacks, like prompt injection, and defenses for Large Language Model (LLM) agents. | | Persona | - | - | | Runtime | - | - | | License | Apache-2.0 | MIT | | Categories | Evaluation & Observability | AI Agents, Evaluation & Observability | ## Trust and health _Sourced signals - not a safety guarantee. No winner column._ | | [jailbreak-evaluation](/tools/controllability-jailbreak-evaluation.md) | [agentdojo](/tools/ethz-spylab-agentdojo.md) | | --- | --- | --- | | Maintenance | Dormant (18%) | Steady (60%) | | Days since push | 638d | 63d | | Open issues (now) | 0 | 41 | | Full report | [trust report](/tools/controllability-jailbreak-evaluation/trust.md) | [trust report](/tools/ethz-spylab-agentdojo/trust.md) | ## Shared compatibility - **Python**: [jailbreak-evaluation](/tools/controllability-jailbreak-evaluation.md) - Python runtime; [agentdojo](/tools/ethz-spylab-agentdojo.md) - Python runtime ## Decision facts: jailbreak-evaluation - **Requirements:** The tool depends on having PyTorch and FastChat installed; An API key from the OpenAI Platform is required for full functionality - **Adopt for:** jailbreak-evaluation is a Python package aimed at evaluating if AI models have been jailbroken by generating outputs that diverge from expected programming. ## Decision facts: agentdojo - **Pricing:** freemium - Open-source under the MIT License. Some advanced features might require additional libraries or APIs. - **Requirements:** Min 8 GB RAM - **Adopt for:** AgentDojo serves as a benchmarking environment to evaluate security attacks, like prompt injection, and defenses for Large Language Model (LLM) agents. ## Choose when ### Choose jailbreak-evaluation if… - License: jailbreak-evaluation is Apache-2.0, agentdojo is MIT. - Requirements: The tool depends on having PyTorch and FastChat installed; An API key from the OpenAI Platform is required for full functionality. - Tags unique to jailbreak-evaluation: ai safety, evaluation tools, jailbreaks, language-models. - When you need to assess whether an AI model can be manipulated to produce unpredictable or unintended outcomes through specific inputs, such as jailbreaking. ### Choose agentdojo if… - License: agentdojo is MIT, jailbreak-evaluation is Apache-2.0. - Pricing: Open-source under the MIT License. Some advanced features might require additional libraries or APIs.. - Requirements: Min 8 GB RAM. - Tags unique to agentdojo: benchmark, large language models, prompt-injection, security. - Also covers AI Agents. - AgentDojo serves as a benchmarking environment to evaluate security attacks, like prompt injection, and defenses for Large Language Model (LLM) agents. ## When NOT to use jailbreak-evaluation - If your project does not involve assessing the security or integrity of how an AI model responds to manipulative input techniques designed to exploit design weaknesses. - When you do not need dependencies on specific frameworks like PyTorch and FastChat, as jailbreak-evaluation requires these without automating their installation. ## When NOT to use agentdojo - AI Agents: Don't use an agent loop when a deterministic workflow would do; agents add latency, cost, and non-determinism. - Evaluation & Observability: Defer heavyweight eval infra only until you have real traffic - never skip it once users depend on answers. ## Common questions ### What is the difference between jailbreak-evaluation and agentdojo? jailbreak-evaluation: Python package for language model jailbreak evaluation. agentdojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents. See the comparison table for live GitHub stats and shared categories. ### When should I choose jailbreak-evaluation over agentdojo? Choose jailbreak-evaluation over agentdojo when License: jailbreak-evaluation is Apache-2.0, agentdojo is MIT; Requirements: The tool depends on having PyTorch and FastChat installed; An API key from the OpenAI Platform is required for full functionality; Tags unique to jailbreak-evaluation: ai safety, evaluation tools, jailbreaks, language-models; When you need to assess whether an AI model can be manipulated to produce unpredictable or unintended outcomes through specific inputs, such as jailbreaking. ### When should I choose agentdojo over jailbreak-evaluation? Choose agentdojo over jailbreak-evaluation when License: agentdojo is MIT, jailbreak-evaluation is Apache-2.0; Pricing: Open-source under the MIT License. Some advanced features might require additional libraries or APIs.; Requirements: Min 8 GB RAM; Tags unique to agentdojo: benchmark, large language models, prompt-injection, security; Also covers AI Agents; AgentDojo serves as a benchmarking environment to evaluate security attacks, like prompt injection, and defenses for Large Language Model (LLM) agents. ### When should I avoid jailbreak-evaluation? If your project does not involve assessing the security or integrity of how an AI model responds to manipulative input techniques designed to exploit design weaknesses. When you do not need dependencies on specific frameworks like PyTorch and FastChat, as jailbreak-evaluation requires these without automating their installation. ### When should I avoid agentdojo? AI Agents: Don't use an agent loop when a deterministic workflow would do; agents add latency, cost, and non-determinism. Evaluation & Observability: Defer heavyweight eval infra only until you have real traffic - never skip it once users depend on answers. ### Is jailbreak-evaluation or agentdojo more popular on GitHub? agentdojo has more GitHub stars (716 vs 27). Stars measure visibility, not whether either tool fits your constraints. ### Are jailbreak-evaluation and agentdojo open source? Yes - both are open-source projects on GitHub (jailbreak-evaluation: Apache-2.0, agentdojo: MIT). ### Where can I find alternatives to jailbreak-evaluation or agentdojo? GraphCanon lists graph-backed alternatives at [jailbreak-evaluation alternatives](/tools/controllability-jailbreak-evaluation/alternatives) and [agentdojo alternatives](/tools/ethz-spylab-agentdojo/alternatives) ([jailbreak-evaluation markdown twin](/tools/controllability-jailbreak-evaluation/alternatives.md), [agentdojo markdown twin](/tools/ethz-spylab-agentdojo/alternatives.md)), ranked by typed relationship edges rather than popularity votes. ### Is there a machine-readable version of this comparison? Yes. The markdown twin at [this comparison](/compare/controllability-jailbreak-evaluation-vs-ethz-spylab-agentdojo.md) mirrors this page for agents and LLM crawlers, with the same stats table and FAQ answers. ### Which is better maintained, jailbreak-evaluation or agentdojo? jailbreak-evaluation: Dormant. agentdojo: Steady. Compare maintenance labels, days since push, and release cadence in the trust section below - stars alone do not measure maintenance. ### Where are the full trust reports for jailbreak-evaluation and agentdojo? GraphCanon publishes per-repo trust reports with dated maintenance, provenance, and scan summaries: [jailbreak-evaluation trust report](/tools/controllability-jailbreak-evaluation/trust); [agentdojo trust report](/tools/ethz-spylab-agentdojo/trust). --- **Machine-readable endpoints** - JSON: [`/api/graphcanon/graph?tool=controllability-jailbreak-evaluation`](/api/graphcanon/graph?tool=controllability-jailbreak-evaluation) - LLM index: [/llms.txt](/llms.txt) - Full corpus: [/llms-full.txt](/llms-full.txt) _GraphCanon - The knowledge graph for AI development. https://www.graphcanon.com/_